Add a LearningCurveSplitter to test sample efficiency

[CompRhys]

Describe the workflow you want to enable

In applications of machine learning the error of the model typically follows a power law with the number of training examples. In the ML for material science and molecules community there is lots of interest in using such plots to asses model performance however as of yet there is no standard splitting tool to make such splits. A standard splitter would improve the reproducibility and uptake of such studies.

Describe your proposed solution

A learning-curve splitter that returns a series of training sets of various sizes alongside a fixed test set. Kwargs would control whether the splitting size was logarithmic, how many splits, whether to take multiple splits for each training set size, etc

Describe alternatives you've considered, if relevant

No response

Additional context

This is an example of the type of plot used to compare models. https://www.nature.com/articles/s41467-020-18556-9